JP2008198157A - Table recognition program, method, and device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To accurately detect ruled lines of a table included in a document image. <P>SOLUTION: A first table recognition program of a table recognition device reading the document image including the table to extract ruled lines is characterized in that the table recognition device is allowed to execute the following processing: run line segment information in the document is detected by run length processing; edge line segment information in the document is detected by edge extraction processing; the kinds of candidates of ruled lines of the table are discriminated according to conditions which predetermine relationships between the adjacent state of the run line segment information and edge line segment information and the kinds of the candidates of the ruled lines; and condition information for discriminating whether to be determined as the ruled lines is switched according to the kind of the candidates of the ruled lines to obtain ruled line information of the table from the candidates of the ruled lines and the condition information. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a table recognition method for extracting ruled lines of a table in a document image.

  There is a document image recognition technique such as OCR (Optical Character Reader) or Optical Character Recognition as a technique for digitizing a business that has been operated with a paper document or converting a document distributed with paper into an electronic document. Documents can contain tables, and the ability to recognize tables is important. A table is generally composed of a combination of vertical and horizontal ruled lines. The table is recognized by extracting the arrangement information of the ruled lines of the table from the document image and analyzing the structure of the table based on the extracted arrangement information of the ruled lines. Therefore, a technique for extracting ruled lines is necessary for accurately recognizing the table.

  As a method for extracting the ruled line of the table, for example, there is a method of detecting a ruled line from continuous pixels of the document image. The method of detecting ruled lines from continuous pixels has good accuracy when detecting solid lines, but cannot detect line segments other than solid lines. On the other hand, there is also a method for detecting ruled lines from a technique for extracting edges in an image. When a ruled line is detected by an edge extraction technique, two ruled line candidates are generated from the solid line and need to be integrated by post-processing. Also, the accuracy is worse than the method of detecting ruled lines from continuous pixels. Even when the ruled lines are detected by both methods and the results of both methods are integrated, post-processing is required. As described above, it is difficult to extract a ruled line from an image in which a plurality of types of ruled lines are mixed by simply combining the method of detecting a ruled line from continuous pixels and the method of detecting a ruled line from the edge extraction technique. there were.

  Moreover, the boundary ruled line formed by the texture cannot be detected by the method of detecting the ruled line from the continuous pixels. On the other hand, the boundary ruled lines formed by texture increase the number of false extractions from other than the ruled lines such as characters in the image in the method of detecting the ruled lines by the technique of extracting edges.

As the prior art, there are the following documents.
Japanese Patent Laid-Open No. 10-40333 JP-A-01-217583

  An object of the present invention is to accurately detect a ruled line of a table represented by a shape, a pattern, and a color included in a document image.

  A first table recognition program of a table recognition apparatus that reads a document image including a table and extracts ruled lines according to the present invention causes the table recognition apparatus to execute the following processing. Run line information in the document is detected by the run length process, edge line information in the document is detected by the edge extraction process, and the run line information and the adjacent state of the edge line segment information and the ruled line The type of ruled line candidate of the table is determined according to a predetermined condition with respect to the type of candidate, the condition information for determining whether or not to use the ruled line is switched according to the type of ruled line candidate, Table ruled line information is obtained from candidates and condition information.

  The second table recognition program of the present invention further causes the table recognition apparatus to execute the following processing. The area where the edge line segment is adjacent to the run line segment is determined as a solid line rule candidate, and the area where the edge line segment is adjacent to one side of the run line segment is determined as a boundary rule line candidate. And a region having a positional relationship in which no run line segment adjacent to the edge line segment exists is determined as a texture boundary candidate. The length condition information for determining a texture boundary candidate as a ruled line is made longer than the condition information about the length of the solid line candidate and the boundary ruled line candidate.

  The third table recognition program according to the present invention further causes the table recognition apparatus to execute the following processing. When the obtained plurality of ruled line information are parallel and less than a predetermined distance, the plurality of ruled line information is integrated to generate new ruled line information.

  The fourth table recognition program according to the present invention further causes the table recognition apparatus to execute the following processing. When performing the step of generating new ruled line information, the condition information is switched according to the solid line rule candidate, the boundary ruled line candidate, and the texture boundary candidate.

  The fifth table recognition program according to the present invention further causes the table recognition apparatus to execute the following processing. Whether ruled line information having a predetermined length or more in ruled line information is defined ruled line information, a rectangular area is specified by a set of the fixed ruled line information adjacent in parallel, and ruled line information in the rectangular area is defined as a ruled line A length condition for determining whether or not is determined is determined by the size of the rectangular area, and ruled line information in the rectangular area exceeding the length condition is defined as a ruled line.

  According to the present invention, the type of the ruled line is specified by the positional relationship between the ruled line candidate by the run detection and the ruled line candidate by the edge detection, and the ruled line is detected according to the condition according to the type of the ruled line. It becomes possible to detect the ruled line of the table represented by the pattern and color with high accuracy.

  Embodiments of the present invention will be described below.

  FIG. 1 is a block diagram showing the principle of this embodiment. When the document image data is input, the table recognition apparatus 10 performs the following processing and outputs ruled line information of the table. The image input means 21 acquires input document image data. The run line segment detection means 22 detects ruled line candidates of the document image data by run length processing. The edge line segment detection means 23 detects ruled line candidates of the document image data by edge detection processing.

  The ruled line candidate extracting unit 24 superimposes the ruled line candidate of the run line segment detecting unit 22 and the ruled line candidate of the edge line segment detecting unit 23, and determines the type of the ruled line candidate according to a predetermined condition. The ruled line information generating means 25 determines whether or not the ruled line information is in accordance with the type of ruled line candidate. The deleting unit 26 deletes inappropriate ruled line information in the table items. The output means 27 outputs ruled line information of the table finally detected from the document image data.

  FIG. 2 is a hardware configuration diagram of the table recognition apparatus 10 according to the present embodiment. The table recognition apparatus 10 includes a control unit 11, an input unit 12, an output unit 13, a storage unit 14, and a memory 15, and each unit is connected by a bus 16. The control unit 11 is a central processing unit that controls the entire table recognition apparatus 10. The control unit 11 is a CPU, for example. The control unit 11 executes the processing of FIG. 1 using the table recognition program 1 and data developed in the memory 15.

  The input unit 12 has a function of receiving document image data to be processed. For example, a scanner, a network interface, a keyboard, a mouse, a touch panel, and the like. The network interface is an interface for the control unit 11 to transmit / receive data to / from an external computer device via a network (for example, the Internet, LAN, etc.). The output unit 13 has a function of outputting ruled lines in the table of document image data. For example, a display device such as a monitor, a printer, or a network interface.

  The table recognition program 1 is stored in the storage unit 14. The storage unit 14 is, for example, a magnetic disk device or a ROM. The memory 15 is an area for temporarily storing the table recognition program 1 and calculation result data stored in the storage unit 14 in order for the control unit 11 to execute the table recognition program 1. The memory 15 is a RAM, for example.

  The control unit 11 expands the table recognition program 1 in the storage unit 14 in the memory 15. According to the table recognition program 1, the control unit 11 serves as an image input unit 21, a run line segment detection unit 22, an edge line segment detection unit 23, a ruled line candidate extraction unit 24, a ruled line information generation unit 25, a deletion unit 26, and an output unit 27. Function.

  The document image data input in this embodiment includes a table. The table recognition apparatus 10 extracts a ruled line of the table. For example, when the document image input to the table recognition device 10 is form data, the table recognition device 10 recognizes a table in the form.

  Next, image input means will be described. The image input means realizes a function of reading document image data into the apparatus. For example, if the image input means reads an image from a paper document, it acquires document image data digitized by an optical scanner. For example, when the document image data already exists, the image input unit acquires the document image data from the storage unit 14 or the like.

  FIG. 3 is a table 30 of document image data input in this embodiment. Table 30 has various forms of ruled lines. A ruled line is represented by a shape, a pattern, a color, or a combination of a shape, a pattern, and a color. The table ruled lines include a boundary ruled line 31 configured by area boundaries, a boundary ruled line 32, a texture boundary ruled line 33 configured by texture area boundaries, a texture boundary ruled line 34, a solid line ruled line 35 configured by solid lines, and a solid line ruled line. 36, a solid line ruled line 37, and a solid line ruled line 38.

  FIG. 4 is a table 40 output from the table 30 in this embodiment. As the table ruled lines, 41, 42, 43, 44, 45, 46, 47, and 48 are extracted as solid lines. Here, since the ruled lines 35, 36, 37, and 38 and the boundary ruled lines 31 and 32 constituted by solid lines are constituted by boundaries of areas (surface-painted areas) filled with the same type of pixels, the run line segment detecting means 22 and the edge line segment extraction means 23 can be detected. On the other hand, the run line segment detection means 22 cannot extract the texture boundary ruled lines 33 and 34. Therefore, it is necessary to use an edge extraction technique in order to extract ruled lines as shown in Table 40 from Table 30 in which a plurality of types of ruled lines are mixed as shown in FIG. In this embodiment, the Canny method will be described as one of the edge extraction techniques.

  Next, processing executed by the run line segment detection unit 22 will be described. The run line segment detection means 22 of this embodiment binarizes each pixel of the document image data with white and black. In the following description, white pixels when binarized are white pixels, and black pixels are black pixels. Run line segment detection is generally called run-length processing. In the run length processing, an area where black pixels are continuous in a vertical direction or a horizontal direction for a predetermined number of pixels or more is extracted as a run line segment area. Therefore, a region in which black pixels are linearly continuous is extracted by the run length process.

  FIG. 5 is a principle diagram of the run length process. In the run length processing, the pixels 6 in the image data are binarized, the same kind of pixel group is extracted, and the line width 50, the start point 51, and the end point 52 formed by the same kind of pixel group are obtained.

  FIG. 6 is a flowchart of processing executed by the run line segment detection means 22. The run line segment detection unit 22 converts the document image data acquired by the image input unit 21 into a binary image (S01). The run line segment detection means 22 detects a region in which black pixels continue in a straight line as a run line segment candidate (S02). The run line segment detection means 22 collects continuous pixels for each row made up of pixels and detects a run line segment candidate in the horizontal direction. The run line segment detection unit 22 deletes the line segment whose length is equal to or less than a predetermined value (S03).

  FIG. 7 is a diagram showing a state of deleting candidate run line segments. 7 shows a state in which the run line segment detection means 22 has extracted the run line segment candidates from the document image data in S02, and the lower side 72 shows the run line segment detection means 22 from the run line segment candidates in S03. Shows the state where the run line segment is deleted. Each round shape in the figure is a pixel 6. The black pixel 6 is a black pixel 601, and the white pixel 6 is a white pixel 602. For example, the run line segment detection unit 22 deletes the run line segment candidates 61 having a length of 3 or less continuous in the horizontal direction 63 of the pixels. As a result, the pixel group 64 corresponding to the run line segment candidate of length 3 near the center of the lower state diagram of FIG. 7 is deleted. Returning to the description of FIG.

  Next, the run line segment detection means 22 groups black pixels (S04). FIG. 8 is a diagram for explaining a grouping state of black pixels. Adjacent run line segment candidates 61 in the document image data are grouped together, and a rectangular area 62 connected to the run line segment candidates 61 is detected. The run line segment detection means 22 sets a rectangular area 62 connected to the run line segment candidates 61 as a ruled line candidate. The solid line ruled line is extracted by the above processing.

  Next, the run line segment detection means 22 determines whether or not it is an area boundary in order to detect a ruled line candidate from the boundary ruled line (S05). Specifically, the run line segment detection means 22 determines that it is an area boundary when the width of the rectangular area of black pixels exceeds a predetermined threshold. The ruled line candidates for the region boundary are the boundary portions at both ends of the rectangular region. In the case of an area boundary (S05: yes), the run line segment detection unit 22 sets both ends of the rectangular area as rule line candidates (S06).

  FIG. 9 shows a setting example of ruled line candidates for the region boundary. The upper part of FIG. 9 is a rectangular area of grouped black pixels. The width 66 of the black pixel line segment is composed of six pixels. The run line detection means 22 generates a region boundary from a group of black pixels. For example, it is assumed that the run line segment detection means 22 has a value of “4” as threshold information for determining whether or not it is an area boundary in advance. The run line segment detection means 22 determines that it is an area boundary when the number of pixels that is the width of the line segment of the black pixel is “4” or more. If it is determined that the region is a boundary, as shown in the lower part of FIG. 9, the boundary between the white pixel and the black pixel at the upper end of the rectangular region is the ruled line candidate 67, and the boundary between the white pixel and the black pixel at the lower end of the rectangular region. Is determined as a ruled line candidate 68.

  In the above processing, the run line segment detection unit 22 detects a ruler line candidate in the horizontal direction. The run line segment detection means 22 also detects ruled line candidates in the vertical direction. The processing executed by the run line segment detection means 22 may be performed by changing the direction of the vertical ruled line and the horizontal ruled line. As described above, the run line segment detecting means 22 extracts the rule line candidates of the run line segments of the solid line border lines and the boundary rule lines.

  Next, the processing of the edge line segment detection means 23 will be described. FIG. 10 is an explanatory diagram of edge extraction.

  In the edge extraction, pixel boundaries 91 on both sides of a straight line 92 formed by black pixels are extracted as lines. In edge extraction, two line segments are extracted on both sides of a straight line 92 formed by one pixel.

  The edge line segment detection means 23 needs to have a function of extracting a texture region boundary. FIG. 11 is an explanatory diagram of the boundary of the texture region 94 detected by edge extraction. It is necessary to detect the boundary line 95 for the texture region 94 in which black pixels are discontinuous. In this embodiment, an edge extraction method called a Canny method is used. The Canny method considers that the fluctuation of the pixel value due to the fine pattern in the texture area is noise superimposed on the surface painting area. First, the input image is smoothed by a Gaussian filter, and this fine pattern is spread to a uniform area to some extent. Thereafter, a change value of the pixel value is obtained by a Sobel filter or the like, and the maximum value of the value is set as an edge pixel. Finally, pixels having a large gradient value in the vicinity of the edge pixel are connected to obtain the edge as a continuous line drawing.

  FIG. 12 is a flowchart of the processing of the edge line segment extraction means using the Canny method. FIG. 13 is a diagram for explaining the processing state of each step in the flowchart of FIG. In this embodiment, the document image data acquired by the image input means 21 is a grayscale image having a minimum value of 0 and a maximum value of 255.

  The edge line segment detection means 23 smoothes between the pixels of the document image data indicated by 1101 in FIG. 13 acquired by the image input means 21 (S11). The image smoothing means uses, for example, a Gaussian filter, and is obtained by the convolution sum of the input image and the Gaussian filter. For example, the input image is I (i, j) and the filter coefficient is F (i, j). FIG. 14 shows an example of the Gaussian filter coefficient 1201. If the image after smoothing conversion is I ′ (i, j), it can be realized by an operation such as the equation (1).

If the height and width of the filter are W (here, “5”), w is “2” from the equation (2).

C is a normalization constant. In this case, a value obtained by adding all the coefficients of each element of the filter of FIG. 14 (C = 159) is used. A state 1102 in FIG. 13 is a state in which the state 1101 is smoothed.

  The edge line segment detection means 23 obtains a gradient value (S12). The gradient value of the corresponding pixel becomes large when the difference in numerical value between the adjacent pixels when smoothing is large. The gradient value is acquired using, for example, a Sobel filter shown. FIG. 15 shows a Sobel filter 1301 for detecting a lateral edge. FIG. 16 shows a Sobel filter 1401 for detecting a vertical edge. For example, when extracting horizontal ruled line candidates, the Sobel filter 1401 for detecting vertical edges in FIG. 16 is used. On the other hand, when extracting vertical ruled line candidates, the Sobel filter 1301 for detecting horizontal edges in FIG. 15 is used. Specifically, the edge line segment detection means 23 calculates each pixel by setting the constant C to “1” using FIGS. 15 and 16 for the coefficient of the filter of Expression (1). There is also a method for obtaining an edge using a Laplacian filter. A state 1103 in FIG. 13 is a state in which the gradient value is acquired from the state 1102. Note that when the state 1103 is viewed from the lateral direction, the state 1104 is obtained.

  The edge line segment detection means 23 acquires the maximum point of the gradient value (S13). For example, when the gradient value of the maximum point is larger than a predetermined threshold value (for example, an arbitrary numerical value such as “1”), the edge line segment detection unit 23 determines that the maximum point is a pixel constituting the edge. Alternatively, the edge line segment detection means 23 is obtained by, for example, obtaining a point whose gradient value is larger than the gradient value of the adjacent point. A state 1105 in FIG. 13 is a state in which the maximum point of the gradient value is acquired from the state 1103.

  The edge line segment detection means 23 detects continuous edges (S14). For example, the edge line segment detection means 23 detects continuous edges by a process generally called a hysteresis process. Specifically, the edge line segment detection means 23 has an undetermined pixel next to a pixel constituting the determined edge, and the gradient value of the undetermined pixel is greater than a predetermined threshold (for example, “2”). If it is larger, it is determined that adjacent uncertain pixels are pixels constituting an edge.

  FIG. 17 is a specific explanatory diagram of pixels constituting an edge by hysteresis processing. The hysteresis processing changes from state 1701 to state 1702, state 1703, and state 1704. In each state, the black pixel 1705 is a pixel determined as a pixel constituting an edge, and the hatched pixel 1706 is an undefined pixel but has a gradient value larger than a predetermined threshold value. The edge line segment detection means 23 determines the hatched pixel 1706 adjacent to the black pixel 1705 in order. In the state 1701, a black pixel 1705 and a hatched pixel 1706 are adjacent to each other in the regions 1707 and 1708. When the black pixel 1705 and the hatched pixel 1706 are adjacent to each other, the edge line segment detection unit 23 sets the hatched pixel 1706 as the black pixel 1705. In the state 1702, the black pixel 1705 and the hatched pixel 1706 are adjacent to each other in the region 1709. The edge line segment detection means 23 sets the hatched pixel 1706 as the black pixel 1705. In the state 1703, the hatched pixels 1706 exist in the areas 1710 and 1711. Since the region 1710 and the region 1711 are not adjacent to the black pixel 1705, they are deleted. The above processing is continued until there is no pixel whose gradient value exceeds a predetermined threshold next to the pixels constituting the edge. When processing is complete, the edge of state 1704 is detected. A state 1106 in FIG. 13 is a state in which a continuous edge from the state 1105 is detected.

  Originally, in the edge detection by the Canny method, when the maximum point of the gradient is obtained, the gradient direction of the gradient is examined, and whether or not the gradient value of the adjacent pixel exceeds a predetermined threshold is determined along the direction orthogonal to the gradient. Investigate.

  However, an object of this embodiment is to obtain vertical and horizontal ruled lines constituting the table. Therefore, the edge line segment detection means 23 may detect the maximum value of the gradient in the vertical direction in the process of extracting the ruler line candidate in the horizontal direction, and then sequentially check the maximum value of the gradient of the adjacent pixels in the horizontal direction. . The opposite is true when extracting vertical ruled lines.

  The edge pixel obtained as described above is regarded as a black pixel, and the edge line segment detecting means 23 groups the black pixels (S15). Next, the edge line segment detection means 23 determines whether or not it is a region boundary in order to detect a ruled line candidate from the boundary ruled line (S16). In the case of an area boundary (S15: yes), the edge line segment detection means 23 sets both ends of the rectangular area as ruled line candidates (S17). Since the processing from S15 to S17 is the same as the description of the run line segment detection means 22, it is omitted.

  Next, the ruled line candidate extraction unit 24 will be described. FIG. 18 is a flowchart of processing executed by the ruled line candidate extraction unit 24.

  The ruler line candidate extraction unit 24 selects the ruled line candidates from the ruled line candidates based on the positional relationship between the run line segment rule line candidates detected by the run line segment detection unit 22 and the edge line segment rule line candidates detected by the edge line segment detection unit 23. Ruled line information that can be determined is extracted. The ruled line candidate extracting unit 24 superimposes the ruled line candidates obtained from the run line segment detecting unit 22 and the edge line segment detecting unit 23, and the positions of the ruled line candidates of the run line segment detecting unit 22 and the ruled line candidates of the edge line segment detecting unit 23 are overlapped. A relationship is acquired (S21).

  The ruled line candidate extraction unit 24 determines whether or not the positional relationship acquired in S21 satisfies a predetermined positional relationship (S22). When the ruled line candidate extracting unit 24 satisfies the positional relationship (S22: yes), the ruled line candidate extracting unit 24 classifies each ruled line information type (S23).

  Here, the positional relationship conditions will be described. FIG. 19 shows the relationship between the position and type of ruled line information determined from the position of the ruled line candidate.

  The ruled line of the original document image data is a solid ruled line 1901. In the case of a solid ruled line 1901, the run line segment detection unit 22 detects a ruled line candidate 1902. Further, the edge line segment detection means 23 detects two ruled line candidates as a ruled line candidate 1903 and a ruled line candidate 1904. The positional relationship when the ruled line candidates are overlapped is a positional relationship in which the ruled line candidate 1903 obtained by the edge line segment detecting unit 23 and the ruled line candidate 1904 sandwich the ruled line candidate 1902 obtained by the run line segment detecting unit 22. When the edge line segment is in a positional relationship across the run line segment, that is, when the edge line segment is adjacent to both sides of the run line segment, the ruled line candidate extraction unit 24 determines that the ruled line candidate is a solid ruled line.

  Next, it is assumed that the ruled line of the original document image data is a boundary ruled line 1905. In the case of the boundary ruled line 1905, the run line segment detecting unit 22 detects a ruled line candidate 1906. Further, the edge line segment detection unit 23 detects ruled line candidates 1907. The positional relationship when the ruled line candidates are overlapped is a positional relationship in which the ruled line candidate 1907 obtained by the edge line segment detecting unit 23 and the ruled line candidate 1906 obtained by the run line segment detecting unit 22 are adjacent to each other. When the edge line segment and the run line segment are adjacent to each other, the ruled line candidate extraction unit 24 determines that the ruled line candidate is a boundary ruled line.

Next, it is assumed that the ruled line of the original document image data is the texture boundary ruled line 1908. In the case of the texture boundary ruled line 1908, the run line segment detecting unit 22 does not detect a ruled line candidate. On the other hand, the edge line segment detection unit 23 detects ruled line candidates 1909.
The positional relationship when the ruled line candidates are overlapped is the positional relationship in which the ruled line candidate 1909 obtained by the edge line segment detection unit 23 exists. In the case of a positional relationship in which only an edge line segment exists, the ruled line candidate extraction unit 24 determines that the ruled line candidate is a texture boundary ruled line.

  If the run line segment length of the rule line candidate by the run line segment detection unit 22 and the length of the edge line segment of the rule line candidate by the edge line segment detection unit 23 are different, the rule line candidate extraction unit 24 determines, for example, the following determination: I do. If the length of the line segment detected by either the run line segment detection means 22 or the edge line segment detection means 23 is more than half the length of the line segment detected by the other, the positional relationship is determined in S22. Alternatively, if the length of the line segment detected by either the run line segment detection means 22 or the edge line segment detection means 23 is less than half the length of the line segment detected by the other, only the longer line segment is selected. To enable.

  Further, when the condition of FIG. 19 is not satisfied, for example, when two run line segments and two edge line segments are adjacent to each other, the ruled line candidate extraction unit 24 performs, for example, run in the adjacent line segments. If there is a line segment, it is determined as a solid ruled line, and if all are edge line segments, it is possible to detect the ruled line by determining as a ruled line candidate from the texture boundary.

  In addition, there are various types of ruled lines such as dotted lines and double lines. Therefore, ruled line candidate type determination can be expanded. For example, when there are three rule line candidates for the edge line and two rule line candidates for the run line, and when the rule line candidates are alternately adjacent to each other, the rule line candidate extracting unit 24 determines the rule line of the document image as a double line. Can be determined. Further, since the edge line segment is easily extracted from the dotted line, the ruled line candidate extracting means 24 is likely to be a dotted line when the edge line segment is extracted alone and the vicinity of the ruled line candidate is a flat region. It is also possible to determine.

  Next, the ruled line information generation means 25 will be described. FIG. 20 is a flowchart of processing executed by the ruled line information generation unit 25. The ruled line information generating unit 25 generates ruled line information from the ruled line candidates. Specifically, deletion of inappropriate ruled line candidates, integration and correction of ruled line candidates, and the like are performed. When the input document image itself is deteriorated, the ruled lines of the table may be affected by blurring or discoloration. If the display of the ruled lines in the table of the original document image is not accurate, the result of extracting the straight lines constituting the ruled lines may be interrupted or deformed in the middle. Therefore, the ruled line information generating unit 25 extracts line segments representing a part or the whole of the straight lines constituting the ruled line as candidates for ruled lines, deletes unnecessary ones of the ruled line candidates, or integrates adjacent ruled line candidates. The ruled line information as the final ruled line extraction result is generated.

  The ruled line information generating unit 25 sets parameters set in advance according to the types of ruled line candidates extracted by the ruled line candidate extracting unit 24 (S31). The parameter is used when generating ruled line information from ruled line candidates. The parameter is, for example, a threshold value for determining whether or not two ruled line candidates exist in parallel and are combined to be regarded as one ruled line. The optimum parameter value varies depending on the type of ruled line candidate. Therefore, the ruled line information generating unit 25 has different parameter values depending on the types of ruled line candidates.

  Examples of parameters of the ruled line information generation unit 25 include the following. The ruled line information generation unit 25 determines whether the ruled line candidate is a ruled line from the relationship between the length information of the ruled line candidate and the threshold value. The parameter sets a threshold value for determining the ruled line candidate from the texture boundary as a ruled line to be twice the threshold value by which the ruled line information generating unit 25 determines the ruled line candidate from the solid line ruled line and the boundary ruled line as the ruled line. In order to be recognized as a texture boundary ruled line by increasing the threshold value for determining the texture boundary as a ruled line, it is necessary to be a straight line longer than the solid line boundary or the boundary ruled line.

  The reason for increasing the threshold value is that the position of the texture boundary ruled line is likely to be ambiguous and noise is likely to occur compared to the solid line ruled line and the boundary ruled line. Furthermore, the reason why the threshold value is increased is that the number of cases where another ruled line exists in the immediate vicinity of the texture boundary is generally smaller than that of the solid line ruled line or the boundary ruled line.

  The ruled line information generation unit 25 can extract an erroneous ruled line and prevent a ruled line that is divided in the middle by switching parameters for detection according to the type of ruled line candidate. The ruled line information generation means 25 determines whether or not a predetermined condition is satisfied (S32). When the ruled line information generation unit 25 determines that the predetermined condition is satisfied (S32: yes), the ruled line information generation unit 25 executes a process according to the condition (S33). Hereinafter, the integration process, the deletion process, and the correction process executed by the ruled line information generation unit 25 will be described.

  Here, a process of generating one piece of ruled line information by integrating the ruled line candidates close by the ruled line information generating unit 25 will be described.

  The ruled line information generation unit 25 determines whether or not a ruled line candidate should be integrated to generate a new ruled line candidate. When a part of the ruled line candidate is deformed, the ruled line width and length of the entire ruled line information are recalculated. The ruled line information generation means 25 converts not only the recalculation of coordinate values but also the attributes such as the coordinates and type of ruled lines into optimum values. For example, when the ruled line information generation unit 25 determines that the ruled line candidates are integrated as one ruled line when the distance between ruled line candidates is short, the threshold value is determined when at least one of the ruled line candidates is a texture boundary ruled line. Increases the threshold value. With this configuration, the ruled line information generation unit 25 can perform adjustment that facilitates the integration of solid line rule candidates or boundary rule line candidates.

  FIG. 21 is a first explanatory diagram of integration of ruled line candidates. If some sections of ruled line candidates overlap, the ruled line candidates are integrated. When the interval d234 between the two ruled line candidates 231 and the ruled line candidate 232 is smaller than the threshold Th1, the ruled line candidate 231 and the ruled line candidate 232 are integrated and converted into one ruled line information 235.

FIG. 22 is a second explanatory diagram of integration of ruled line candidates.
The ruled line candidate 241 and the ruled line candidate 242 are in a positional relationship arranged in series instead of in parallel. When the interval d243 between the two ruled line candidates 241 and the ruled line candidate 242 is smaller than the threshold Th2, they are integrated and converted into one ruled line information 244.

FIG. 23 is a third explanatory diagram of integration of ruled line candidates.
The ruled line candidate 251 and the ruled line candidate 252 are arranged in parallel. In this embodiment, two types of threshold values Th3 and Th4 are used. The length of the short ruled line candidate 251 of the ruled line candidates 251 and 252 is L1, and the length of the long ruled line candidate 252 is L2. When the distance d253 between the two ruled line candidates 251 and the ruled line candidate 252 is smaller than the threshold Th3, and the ratio of the length L2 of the ruled line candidate 252 and the length L1 of the ruled line candidate 251 is larger than the product of the threshold Th4. Both are integrated and converted into a piece of ruled line information 254. Specifically, the ruled line information generation unit 25 deletes the ruled line candidate 251. The relationship between the length L1 of the ruled line information 251 and the length L2 of the ruled line information 252 is such that L2 is longer than L1 so that L1 can be regarded as noise of L2.

  For example, when the input image is about 200 dpi, Th1 = 8 dots (about 0.1 mm), Th2 = 16 dots (about 0.2 mm), Th3 = 8 dots (about 0.1 mm), and Th4 = 5 dots (about 0.06 mm). ) Etc.

  Further, as another criterion for deleting ruled line candidates, the ruled line information generating unit 25 can delete a ruled line candidate whose length does not reach the threshold value. Even when it is not adjacent to other ruled line candidates, the ruled line information generating unit 25 deletes ruled line candidates whose length is shorter than a predetermined threshold. For example, when the threshold is set to 20, ruled line candidates whose ruled line candidate length is less than about 2.54 mm are deleted at 200 dpi. In general, the length of the ruled lines constituting the table in the document image data has a certain lower limit, and the above threshold values are the ruled line candidates erroneously extracted from the characters and the ruled line candidates extracted from the ruled lines. Can be distinguished from each other.

  The ruled line information generating unit 25 changes the attributes such as the position and size of the ruled line information from the ruled line candidates. In the case of changing, the ruled line information generating means 25 has a function of discriminating whether or not there is a change depending on the type of whether the ruled line candidate including the pixel to be corrected is a solid line ruled line, a boundary ruled line or a texture boundary ruled line. For example, in the case of a texture boundary ruled line candidate, the ruled line information generation unit 25 checks whether the area in contact with the texture boundary ruled line candidate is a texture area. Only when one of the areas in contact with the texture boundary ruled line candidate is a texture area, or when both areas in contact with the texture boundary ruled line candidate are different types of texture areas, it is determined to be determined as a ruled line candidate. It is possible. This will be specifically described below.

FIG. 24 is an explanatory diagram of texture boundary determination.
A state 2601 includes a texture region 2603 and a white region 2604, and a ruled line candidate 2602 at the texture boundary is detected by the ruled line candidate extraction unit 24.

  A state 2605 is a state in which there is a character string 2607 in the white area, and the edge detection unit 23 erroneously detects a ruled line candidate 2606 for the texture boundary ruled line at the lower end of the character string 2607. This is because the edge detection means 23 may misrecognize that it is a texture boundary because the lower end of a horizontally written character string is aligned on the horizontal axis. This is because the envelope of the character string is not a texture boundary, but the image feature of the envelope portion of the character string and the image feature indicating the texture boundary are very similar.

  Therefore, the ruled line information generation unit 25 checks whether the area between the ruled lines is a texture area, and determines whether it is a ruled line candidate obtained from a texture boundary or a ruled line candidate obtained from an envelope of a character string.

  It should be noted that the determination of whether or not the texture region is present cannot be made when there is a solid line at the position of the ruled line candidate, only by examining the area around the ruled line candidate. However, in the present embodiment, the ruled line candidate extracting unit 24 determines whether the ruled line candidate is any one of a solid line ruled line, a boundary ruled line, and a textured boundary ruled line. Judgment is possible.

As a determination of whether or not it is a texture region, for example, there is a method in which black pixels are grouped for each region on both sides of a ruled line candidate, and determination is made based on an average value of the sizes of black pixel groups. The size of the black pixel group in the texture area is smaller than the size of the black pixel group in the character string. For this reason, if the statistics of the size of the group of black pixels in the case of a character are measured in advance and a threshold value for the size of the group of black pixels that separates the character and the texture is provided, the texture region and the character string can be distinguished.
In FIG. 24, the average value of the black pixel group size 2609 of the texture area in the state 2601 is 8 dots, the average value of the black pixel group size 2610 in the white area of the state 2601 is 0 dots, and the characters in the state 2605 The average value of the size 2611 of the black pixel group in the white area including the column is 100 dots, and the average value of the size 2612 of the black pixel group in the texture area in the state 2605 is 2 dots. Here, it is assumed that the size of the black pixel group is predetermined as “50” as a threshold for determining whether or not it is a texture region.

  The ruled line information generation unit 25 compares the size of the black pixel group included in the areas 2603 and 2604 sandwiching the ruled line candidate 2602 of the texture boundary ruled line in the state 2601 with a threshold value. Sizes 2609 and 2610 are smaller than the threshold value. Therefore, the ruled line information generation unit 25 determines that the ruled line candidate 2602 is a texture boundary ruled line. On the other hand, the ruled line information generating unit 25 compares the size of the area including the character string 2607 sandwiching the ruled line candidate 2606 of the texture boundary ruled line in the state 2605 and the size of the black pixel group included in the area 2604 with the threshold value. The size 2611 is a value larger than the threshold value. Therefore, the ruled line information generation unit 25 determines that the ruled line candidate 2606 is a line in which the lower end of the character string is erroneously detected.

  Next, the correction process executed by the ruled line information generation unit 25 will be described. It is also possible to correct a ruled line candidate when the width and height of the ruled line satisfy a predetermined condition. The correction is a process in which, for example, the ruled line information generating unit 25 sets the width of a ruled line candidate that becomes a border ruled line or a ruled line candidate that becomes a textured border ruled line to a minimum value (for example, 1 dot).

  The predetermined conditions include the following examples. For example, the ruled line obtained from the region boundary does not have a logical width, but the ruled line width is generated in the process of actually extracting the ruled line. Therefore, it is possible to execute a process for correcting the ruled line candidate determined to be a boundary ruled line. As another example, in the black pixel grouping process executed by the run line segment detection unit 22, the width of the ruled line may be increased by the black pixel rectangular area 62 in FIG. Therefore, it is possible to execute processing for correcting the width of the ruled line whose width is increased.

  As described above, the ruled line information generation unit 25 determines the ruled line information based on the positional relationship of the ruled line candidates before being converted into the ruled line information, and determines the ruled line information using the parameters according to the types of the ruled line candidates. To do. As a result, the accuracy of extracting ruled line information is improved.

  Next, a method for deleting ruled line candidates that are erroneously extracted other than ruled lines will be described. The erroneous extraction is to extract line segment information that is not a ruled line such as characters in the image as a ruled line by mistake. If various types of ruled lines such as a solid line ruled line, a boundary ruled line, and a texture boundary ruled line are extracted, the influence of erroneous extraction increases.

  The deleting unit 26 needs to determine a ruled line candidate erroneously extracted from other than the ruled line as an illegal ruled line candidate. The ruled line candidates extracted in error are characters in the table items. FIG. 25 is a diagram showing the relationship between the table 283 and the size of the character string in the table item. The character string 280 inside the table 283 generally fits in the item area constituting the table. Therefore, the length 284 in the vertical direction of the ruled line candidate in which the character string 280 in the item of the table 283 is erroneously extracted is shorter than the height 281 of the area of the item constituting the line. Since the ruled lines in Table 283 are generally connected above and below the rows, they are longer than the heights of the regions constituting the rows. The same applies to the width direction 283 of the items in Table 283. For example, the relationship between the sizes of the other items 285, 286, and 287 in the table 283 and the character strings written in the other items 285, 286, and 287 is the same. The size of the item area in the height direction is larger than the size of the character string in the height direction. Therefore, by specifying the area of each item constituting the table 283, it is possible to determine an appropriate length threshold for deleting ruled line candidates in the item based on the height information or width information of the item. is there. In the following description, the area for determining whether or not to delete ruled line candidates is referred to as a ruled line determination area.

  Next, processing executed by the deleting unit 26 will be described. FIG. 26 is a flowchart of processing executed by the deleting unit 26. The deleting unit 26 sets an area for determining whether or not to delete the ruled line (S41).

  The deleting unit 26 detects a confirmed ruled line in the document image data. As a criterion for determining a confirmed ruled line, for example, a ruled line longer than a predetermined threshold is determined as a confirmed ruled line. For example, when the input document image data has a resolution of 200 dpi, the threshold value used as the determination rule for the confirmed ruled line may be about 100 dots (about 12.7 mm). Setting a long ruled line as a confirmed ruled line can prevent inappropriate ruled line information from being used for setting a ruled line determination area. Inappropriate ruled line information is ruled line information extracted from other than ruled lines such as ruled lines erroneously detected from characters. The deletion unit 26 detects a set of confirmed ruled lines that are parallel and adjacent from the set of confirmed ruled lines, and generates a ruled line determination region.

  FIG. 27 shows a setting example of the ruled line determination area. The ruled line determination area in this embodiment is a rectangular area sandwiched between adjacent long ruled lines. In the following description, a ruled line for specifying a ruled line determination area is defined as a confirmed ruled line. 27 includes long ruled line information 2901, 2902, 2903, and 2904 in the horizontal direction. The table 2900 has 2905 as the vertical ruled line. Regarding the horizontal ruled lines in Table 2900, the sets of ruled line information that are parallel and adjacent to each other among the ruled line information are 2901 and 2902, 2902 and 2903, and 2903 and 2904. The areas sandwiched between the groups are ruled line determination areas 2905, 2906, and 2907 as shown in the lower part of FIG.

  Note that the set of parallel confirmed ruled lines is not necessarily the same length as shown in FIG. For example, there are cases where a set of confirmed ruled lines is partially parallel.

  FIG. 28 is an explanatory diagram showing a state where the confirmed ruled lines in the document image data are partially parallel. In FIG. 28, there are fixed ruled lines 311, 312 and 313. The confirmed ruled lines 311, 312 and 313 are partially parallel to each other. The confirmed ruled lines 311 and 312 are adjacent in parallel within a range 316. The confirmed ruled lines 311 and 313 are adjacent in parallel within a range 317. A rectangular area formed by the determined ruled lines 311 and 312 in the range 316 is a ruled line determination area 314. A rectangular area formed by the determined ruled lines 311 and 313 in the range 317 is a ruled line determination area 315. When a part of the confirmed ruled line is parallel, the deleting unit 26 registers only the part adjacent in parallel as the ruled line determination area.

  Next, processing executed by the deleting unit 26 will be described. FIG. 29 is a flowchart of processing in which the deleting unit 26 sets the specification of the ruled line determination area. The deleting unit 26 specifies an arbitrary confirmed ruled line as a detection target (S51). In FIG. 28, the deleting unit 26 specifies the confirmed ruled line 311 as a detection target. The deletion unit 26 detects a confirmed ruled line that is below the document image data and adjacent in parallel in the horizontal direction with respect to the confirmed ruled line to be detected (S52). In FIG. 28, the deleting unit 26 detects a confirmed ruled line 312 that is below the confirmed ruled line 311 and that is adjacent in parallel in the horizontal direction. When the deleting unit 26 detects an adjacent confirmed ruled line (S52: yes), the deleting unit 26 specifies a ruled line determination region in a range formed from the determined ruled line adjacent to the detection target ruled line (S53). In FIG. 28, when the deletion means 26 detects the confirmed ruled line 312 (S52: yes), it specifies the ruled line determination area 314 created by the range 316 (S53).

  When the deletion unit 26 has not performed the detection process for all the horizontal ranges of the determined ruled lines to be detected identified in S51 (S54: no), the deletion unit 26 detects again the horizontal ranges of the remaining fixed ruled lines to be detected. Process. In FIG. 28, when the deletion unit 26 has not performed detection processing for all the horizontal ranges of the confirmed ruled line 311 (S54: no), the deletion unit 26 performs detection processing again for the horizontal ranges of the remaining confirmed ruled lines 311. In FIG. 28, the deletion unit 26 detects a confirmed ruled line 313 that is below the confirmed ruled line 311 in the range 317 and that is adjacent to the range 316 in parallel in the horizontal direction. In FIG. 28, the deleting unit 26 specifies the ruled line determination area 315 created by the range 317.

  On the other hand, when the deletion unit 26 has performed detection processing for all the horizontal ranges of the confirmed ruled lines to be detected (S54: yes), detection of adjacent confirmed ruled lines for all the confirmed ruled lines in the document image data is completed. It is determined whether or not (S55). In FIG. 28, the deletion unit 26 determines whether or not the detection of adjacent confirmed ruled lines for all confirmed ruled lines in the document image data has been completed when the detection processing is performed for all the horizontal ranges of the confirmed ruled lines 311. To do. When the deletion unit 26 performs the above processing for all confirmed ruled lines specified by the document image data (S55: yes), the ruler determination area is registered and the process is completed.

  Returning to the description of FIG. The deleting unit 26 calculates a ruled line determination value corresponding to the ruled line determination area (S42). The ruled line determination value is a threshold value for determining whether or not the ruled line included in the ruled line determination area is the true ruled line information. In this embodiment, the ruled line determination value is ruled line length information. The deleting unit 26 deletes an invalid ruled line by the ruled line candidate deleting unit based on the length information for determining the ruled line.

  The deleting unit 26 sets a length threshold value for each ruled line determination area. For example, the height information of the ruled line determination area is acquired, and a length slightly shorter than the height information is set as a threshold value. For example, in the case of an image of 200 dpi, the length is about 20 dots shorter than the number of dots in the height of the ruled line determination area. Alternatively, for example, the frequency distribution of the lengths of the ruled line candidates in the ruled line determination area is obtained and the maximum value of the frequency distribution is set as a threshold value, or the length of the ruled line candidate corresponding to the maximum value of the frequency distribution is set as a threshold value There is a method to do.

  Next, the deleting unit 26 deletes inappropriate ruled line information (S43). Specifically, the deleting unit 26 deletes inappropriate ruled line information in the ruled line determination area configured by parallel confirmed ruled lines. The inappropriate ruled line information is ruled line information having a length shorter than the length information for determining the ruled line calculated in S42. In the present embodiment, inappropriate vertical ruled line information in the ruled line determination area constituted by a set of horizontal confirmed ruled line information is deleted. In the process of S43, the ruled line information may be processed to delete not only the vertical ruled line information but also the horizontal ruled line information. Inappropriate ruled line information is often misdetected character information. This is because the erroneously detected characters have substantially the same length in the horizontal direction and the vertical direction.

FIG. 30 is an explanatory diagram when the deleting unit 26 deletes an inappropriate ruled line. The upper table 3401 in FIG. 30 is the ruled line information of the table in the document image data generated by the ruled line information generating unit 25. The table 3401 includes a confirmed ruled line 3402, a confirmed ruled line 3403, a confirmed ruled line 3404, a confirmed ruled line 3405, a confirmed ruled line 3406, and a confirmed ruled line 3407. Further, the table 3401 includes a ruled line determination area 3408 from the confirmed ruled line 3402 and the confirmed ruled line 3403, a ruled line determination area 3409 from the confirmed ruled line 3403 and the confirmed ruled line 3404, and a ruled line determination area 3410 and a confirmed ruled line 3403 from the confirmed ruled line 3404 and the confirmed ruled line 3405. And a ruled line determination area 3411 from the confirmed ruled line 3406, a ruled line determination area 3412 from the confirmed ruled line 3406 and the confirmed ruled line 3407, and a ruled line determination area 3413 from the confirmed ruled line 3407 and the confirmed ruled line 3405.
A table 3420 in the middle of FIG. 30 shows the height of an area for determining a length for deleting inappropriate ruled line information determined for each ruled line determination area. The length information for deleting inappropriate ruled line information determined for each ruled line determination area is calculated by the deleting unit 26 based on the following. The ruled line determination area 3408 is calculated based on the height 3414 of the ruled line determination area 3408. The ruled line determination area 3409 is calculated based on the height 3415 of the ruled line determination area 3409. The ruled line determination area 3410 is calculated based on the height 3416 of the ruled line determination area 3410. The ruled line determination area 3411 is calculated based on the height 3417 of the ruled line determination area 3411. The ruled line determination area 3412 is calculated based on the height 3418 of the ruled line determination area 3412. The ruler determination area 3413 is calculated by the deletion unit 26 based on the height 3419 of the ruled line determination area 3413.

  The deletion means 26 determines whether the ruled line information in each ruled line determination area is appropriate or inappropriate based on the length information for determining the ruled line determined for each ruled line determination area. Specifically, ruled line information shorter than length information for determining a ruled line obtained for each ruled line determination area is deleted. A table 3420 in the lower part of FIG. 30 shows a state in which the ruled line information 3402 in each area of the table 3401 is deleted.

  When the ruled line generating means 25 sets the parameters, it is possible to specify a ruled line determination area and set an appropriate parameter for each ruled line determination area to perform highly accurate ruled line extraction. For example, in the case of a texture region, the threshold value for determining a ruled line may be lengthened.

  The output means 27 outputs the ruled line information obtained as described above.

  As described above, according to this embodiment, even when a plurality of types of ruled lines including solid lines, boundary ruled lines, and texture boundary ruled lines are mixed in the input image, appropriate ruled line extraction processing can be performed according to the respective ruled line types. The accuracy of ruled line extraction can be improved. As a result, the burden of error correction work for ruled line extraction can be reduced, and the number of man-hours for the user can be reduced.

  In addition, since it is possible to change threshold information for deleting inappropriate ruled lines for each area constituting the table, it is possible to prevent erroneous detection even when the size of the items constituting the table is different. Become.

  Conventionally, run line segments and edge line segments are extracted to generate ruled line candidates, and noise is removed from the ruled line candidates to generate ruled line information, and finally obtained results are integrated. That is, the run line segment and the edge line segment are not directly compared. When a texture boundary and a solid line boundary are detected, in edge extraction, a total of three rule line candidates are detected as one line segment candidate, one at the texture boundary and two at the solid line boundary. However, when the texture boundary and the solid line boundary are close to each other, the interval between the edge line segments is close, so that it is difficult to associate the line segments constituting the solid line. Therefore, the prior art cannot determine which of the detected three edge line segments should be integrated and converted to a solid line. In addition, as a conventional technique, a method of executing the run line segment extracting unit and the edge line segment extracting unit in parallel and integrating the respective ruled line extraction results may be considered, but when competing ruled line candidates are extracted from the same region It is still necessary to make difficult decisions, such as having to choose one of them. As described above, it is impossible to extract a ruled line from an image in which a plurality of types of ruled lines are mixed with high accuracy only by combining conventional techniques.

  On the other hand, in this embodiment, the run line segment and the edge line segment are overlapped, and the run line segment is sandwiched between two edge line segments to be integrated into one solid line rule line. And boundary ruled lines can be generated correctly. As described above, a plurality of types of ruled lines can be extracted with high accuracy by comparing the run line segment and the edge line segment before generating the ruled line information from the ruled line candidates. Furthermore, since the ruled line generation parameter is changed according to the type of ruled line candidate, the ruled line can be extracted with high accuracy.

It is a principle block diagram of a present Example. It is a hardware block diagram of the table | surface recognition apparatus 10 of a present Example. It is the table | surface 30 of the document image data input in a present Example. It is Table 40 output from Table 30 in a present Example. It is a principle diagram of run length processing. 4 is a flowchart of processing executed by a run line segment detection unit 22; It is a figure which shows the deletion state of the candidate for a run line segment. It is a figure explaining the state of grouping of black pixels. It is an example of setting a ruled line candidate for an area boundary. It is explanatory drawing of edge extraction. It is explanatory drawing of the boundary of the texture area | region 94 detected by edge extraction. It is a flowchart of the process of the edge line segment extraction means using the Canny method. It is a figure explaining the state of the process of each step in the flowchart of FIG. It is an example of a Gaussian filter coefficient 1201. This is a Sobel filter 1301 for detecting a lateral edge. This is a Sobel filter 1401 for detecting a vertical edge. It is specific explanatory drawing of the pixel which comprises the edge by a hysteresis process. It is a flowchart of the process which the ruled line candidate extraction means 24 performs. This is the relationship between the position and type of ruled line information determined from the position of the ruled line candidate. It is a flowchart of the process which the ruled line information generation means 25 performs. It is a 1st explanatory view of integration of a ruled line candidate. It is a 2nd explanatory drawing of integration of a ruled line candidate. It is a 3rd explanatory drawing of integration of a ruled line candidate. It is explanatory drawing of determination of a texture boundary. It is a figure which shows the relationship between the magnitude | size of the character string in the item of Table 283 and a table | surface. 5 is a flowchart of processing executed by a deleting unit 26. It is a setting example of a ruled line determination area. It is explanatory drawing which shows the state in which the fixed ruled line in document image data is partially parallel. 10 is a flowchart of a process in which deletion means 26 sets identification of a ruled line determination area. It is explanatory drawing when the deletion means 26 deletes an inappropriate ruled line.

Explanation of symbols

21 Image input means 22 Run line segment detection means 23 Edge line segment detection means 24 Rule line candidate extraction means 25 Rule line information generation means 26 Deletion means 27 Output means 10 Table recognition apparatus 10
11 Control unit 11
12 Input means 12
13 Output means 13
14 storage means 14
15 memory 15
16 Bus 16

Claims (7)

A table recognition program for a table recognition device that reads a document image including a table and extracts ruled lines,
Detecting run line information in the document by run length processing in the table recognition device;
Detecting edge line segment information in the document by edge extraction processing;
Discriminating the types of ruled line candidates in the table according to a predetermined condition for the relationship between the adjacent state of the run line segment information and the edge line segment information and the types of ruled line candidates;
Switching condition information for determining whether or not to use a ruled line according to the type of candidate for the ruled line;
Obtaining ruled line information of the table from the ruled line candidates and the condition information;
Table recognition program characterized by causing A region in which the edge line segment is adjacent to each other so as to sandwich the run line segment is determined as a solid ruled line candidate, and a region in which the edge line segment is adjacent to one side of the run line segment is determined as a boundary ruled line. Determining the candidate, and causing the table recognition apparatus to execute a step of determining a region having a positional relationship in which no run line segment adjacent to the edge line segment exists as a texture boundary candidate,
2. The table recognition program according to claim 1, wherein the length condition information for determining the texture boundary candidate as a ruled line is longer than the solid line rule candidate and the length condition information of the boundary ruled line candidate.   The table recognizing apparatus executes the step of generating new ruled line information by integrating the plurality of ruled line information when the obtained ruled line information is parallel and equal to or less than a predetermined distance. The table recognition program according to 1.   The table recognition apparatus, when performing the step of generating the new ruled line information, causes the table recognition apparatus to execute a step of switching the condition information according to the solid line ruled line candidate, the boundary ruled line candidate, and the texture boundary candidate. Item 4. The table recognition program according to Item 3. A step of setting ruled line information longer than a predetermined length in the ruled line information as confirmed ruled line information;
Identifying a rectangular region by a set of the determined ruled line information adjacent in parallel;
Determining a length condition for determining whether or not the ruled line information in the rectangular area is a ruled line according to the size of the rectangular area;
The table recognition program according to claim 1, wherein the table recognition apparatus executes the step of setting ruled line information in the rectangular area exceeding the length condition as a ruled line. A table recognition method for reading a document image including a table and extracting ruled lines,
Computer
Run line processing detects run line information in the document,
Detect edge line segment information in the document by edge extraction processing,
Discriminating the types of ruled line candidates in the table according to a predetermined condition of the relationship between the adjacent state of the run line segment information and the edge line segment information and the types of ruled line candidates;
Switching the condition information for determining whether or not to use a ruled line according to the type of candidate for the ruled line,
A table recognition method, characterized in that ruled line information of the table is obtained from the ruled line candidates and the condition information. A table recognition device that reads a document image including a table and extracts ruled lines,
Run line processing detects run line information in the document,
Detect edge line segment information in the document by edge extraction processing,
Discriminating the types of ruled line candidates in the table according to a predetermined condition of the relationship between the adjacent state of the run line segment information and the edge line segment information and the types of ruled line candidates;
Switching the condition information for determining whether or not to use a ruled line according to the type of candidate for the ruled line,
A table recognition apparatus for obtaining ruled line information of the table from the ruled line candidates and the condition information.